Online Transfer Learning for Concept Drifting Data Streams
Online Transfer Learning for Concept Drifting Data Streams
复制标题
DOI：
--
复制
发表时间：
2019-08
期刊：
影响因子：
--
通讯作者：
Helen McKay;N. Griffiths;Phillip Taylor;T. Damoulas;Zhou Xu
中科院分区：
文献类型：
--
作者：
Helen McKay;N. Griffiths;Phillip Taylor;T. Damoulas;Zhou Xu
迁移学习使用在源域中学习的知识来辅助目标域中的预测。当源域和目标域都在线时，每个域都容易受到概念漂移的影响，这可能会改变它们之间的知识映射。在线领域的漂移可以提供更多的信息，需要从源到目标的知识转移，反之亦然。为了解决这个问题，我们引入了双向在线迁移学习框架（BOTL），该框架使用在每个在线领域学到的知识来帮助其他领域的预测。我们还介绍了两种变体的BOTL，将模型剔除，以尽量减少负转移的框架中有大量的域。我们提供了一个理论上的性能保证，表明BOTL实现的损失不差于底层的本地概念漂移检测算法。使用两个数据流发生器：漂移超平面仿真器和智能家居加热模拟器，以及从车辆遥测预测碰撞时间（TTC）的真实数据，给出了实证结果。评估显示BOTL及其变体优于现有的最先进的技术。
Transfer learning uses knowledge learnt in a source domain to aid predictions in a target domain. When both source and target domains are online, each are susceptible to concept drift, which may alter the mapping of knowledge between them. Drifts in online domains can make additional information available, necessitating knowledge transfer both from the source to the target and vice versa. To address this we introduce the Bi-directional Online Transfer Learning framework (BOTL), which uses knowledge learnt in each online domain to aid predictions in others. We also introduce two variants of BOTL that incorporate model culling to minimise negative transfer in frameworks with large numbers of domains. We provide a theoretical performance guarantee that indicates BOTL achieves a loss no worse than the underlying local concept drift detection algorithm. Empirical results are presented using two data stream generators: the drifting hyperplane emulator and the smart home heating simulator, and real-world data predicting Time To Collision (TTC) from vehicle telemetry. The evaluation shows BOTL and it’s variants outperform the existing state-of-the-art technique.